Deadlock using FactoryModuleBuilder from different threads

[Antony Stubbs]

Hi, we're experiencing a deadlock in our multi-threaded code, and I think I've tracked it down to our use of FactoryModuleBuilder to create one of our objects during an HTTP response. We use the pattern of:

install(new FactoryModuleBuilder()

                .implement(User.class, User.class)

                .build(UserFactory.class));

And our dead lock looks like (we have more than a dozen threads all blocked in the same way):

"Thread-2616" prio=10 tid=0x00007f41647e1800 nid=0xea9 waiting for monitor entry [0x00007f417fefd000]

   java.lang.Thread.State: BLOCKED (on object monitor)

        at com.google.inject.internal.InternalInjectorCreator.build(InternalInjectorCreator.java:102)

        - waiting to lock <0x0000000608d609a0> (a com.google.inject.internal.InheritingState)

        at com.google.inject.internal.InjectorImpl.createChildInjector(InjectorImpl.java:217)

        at com.google.inject.internal.InjectorImpl.createChildInjector(InjectorImpl.java:224)

        at com.google.inject.assistedinject.FactoryProvider2.getBindingFromNewInjector(FactoryProvider2.java:602)

        at com.google.inject.assistedinject.FactoryProvider2.invoke(FactoryProvider2.java:625)

        at $Proxy16.create(Unknown Source)

        at ourCode

So I am concluding that the Factory Module Builder pattern isn't thread safe per-se, and will just go back to creating our object manually, then using injector#injectMembers() afterwards.

Am I missing something or does this sound about right? Is this correct? Or am I missing something? If it is, it would be great to get a warning added somewhere..

The relavent Guice code is FactoryProvider2:

 /**

   * When a factory method is invoked, we create a child injector that binds all parameters, then

   * use that to get an instance of the return type.

   */

  public Object invoke(Object proxy, final Method method, final Object[] args) throws Throwable {

and then the synchro block in InternalInjectorCreator:

  public Injector build() {

    if (shellBuilder == null) {

      throw new AssertionError("Already built, builders are not reusable.");

    }

    // Synchronize while we're building up the bindings and other injector state. This ensures that

    // the JIT bindings in the parent injector don't change while we're being built

    synchronized (shellBuilder.lock()) {

      shells = shellBuilder.build(initializer, bindingData, stopwatch, errors);

      stopwatch.resetAndLog("Injector construction");

      initializeStatically();

    }

    injectDynamically();

    if (shellBuilder.getStage() == Stage.TOOL) {

      // wrap the primaryInjector in a ToolStageInjector

      // to prevent non-tool-friendy methods from being called.

      return new ToolStageInjector(primaryInjector());

    } else {

      return primaryInjector();

    }

  }

[Stuart McCulloch]

Hi Antony,

Could you open an issue and attach the full thread-dump (or at least the 'deadlocks' section that summarises the threads and locks involved in the deadlock).

--

Cheers, Stuart

--
You received this message because you are subscribed to the Google Groups "google-guice" group.
To view this discussion on the web visit https://groups.google.com/d/msg/google-guice/-/BgAsyOu14cIJ.
To post to this group, send email to google...@googlegroups.com.
To unsubscribe from this group, send email to google-guice...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/google-guice?hl=en.

[Antony Stubbs]

As requested:

Issue 729: Deadlock using FactoryModuleBuilder from different threads

http://code.google.com/p/google-guice/issues/detail?id=729